Smart Fraud Detection in Online Retail: Leveraging Classification Algorithms for E-Commerce Transactions

Abstract

The quick increase in e-commerce transactions through the internet has turned fraud detection into a major obstacle to e-commerce operations' security and reliability. This paper suggests a good DNN-based framework for intelligent fraud detection, which is built on the analysis of a very imbalanced credit card fraud dataset. This process consists of numerous data preprocessing steps, which involve, but are not limited to, how the missing values are treated and how duplicates and outliers are eliminated, as well as how z-score normalization can be used to obtain a homogeneous scale. To address the problem of class imbalance, SMOTE is applied, and a minority fraud is oversampled, and Principal Component Analysis (PCA) is applied, and the most beneficial attributes are obtained. The improved data is divided into training and testing (80:20) before training the DNN model. The results of the experiment indicate that the given DNN is more advantageous compared to traditional machine learning algorithms, including ANN, XGBoost, and Decision Tree in terms of accuracy, precision, recall, and F1-score of 97.10%. Overall, the proposed DNN-based system is effective in understanding the complex fraudulent activities in online shopping, it is also highly accurate in detection and offers a scalable solution that can be used to manage the contemporary e-commerce fraud prevention.